1 So, how much of the Earth’s surface is covered by rain gauges? 2 3 By Chris Kidd, Andreas Becker, George J. Huffman, Catherine L. Muller, Paul Joe, Gail Skofronick- 4 Jackson and Dalia B. Kirschbaum 5 6 Affiliations: 7 Kidd - University of Maryland, College Park, Maryland, 20740 and NASA/Goddard Space Flight 8 Center, Greenbelt, Maryland, 20771 9 Becker - Deutscher Wetterdienst, Offenbach am Main, Germany 10 Huffman, Skofronick-Jackson and Kirschbaum - NASA/Goddard Space Flight Center, Greenbelt, 11 Maryland, 20771; 12 Muller - Royal Meteorological Society, Reading, United Kingdom and School of Geography, Earth and 13 Environmental Sciences, University of Birmingham, UK; 14 Joe - Environment Canada, Meteorological Research Division, Toronto, Canada 15 16 Corresponding author: 17 Chris Kidd, Code 613, NASA/Goddard Space Flight Center, Greenbelt, Maryland, 20771. E-mail: 18 [email protected] 19 20 Summary Capsule: 21 The total area measured globally by all currently available rain gauges is surprisingly small, equivalent 22 to less than half a football field or soccer pitch. 23 1 24 Abstract 25 The measurement of global precipitation, both rainfall and snowfall, is critical to a wide range of users 26 and applications. Rain gauges are indispensable in the measurement of precipitation, remaining the de 27 facto standard for precipitation information across the Earth’s surface for hydro-meteorological 28 purposes. However, their distribution across the globe is limited: over land their distribution and 29 density is variable, while over oceans very few gauges exist and where measurements are made, they 30 may not adequately reflect the rainfall amounts of the broader area. Critically, the number of gauges 31 available, or appropriate for a particular study, varies greatly across the Earth due to temporal sampling 32 resolutions, periods of operation, data latency and data access. Numbers of gauges range from a few 33 thousand available in near real time, to about a hundred thousand for all ‘official’ gauges, and to 34 possibly hundreds of thousands if all possible gauges are included. Gauges routinely used in the 35 generation of global precipitation products cover an equivalent area of between about 250 m2 and 3,000 36 m2. For comparison, the center circle of a soccer pitch or tennis court is about 260 m2. Although each 37 gauge should represent more than just the gauge orifice, auto-correlation distances of precipitation vary 38 greatly with regime and the integration period. Assuming each Global Precipitation Climatology Centre 39 (GPCC) -available gauge is independent and represents a surrounding area of 5 km radius, this 40 represents only about 1% of the Earth’s surface. The situation is further confounded for snowfall which 41 has a greater measurement uncertainty. 42 43 2 44 Precipitation, including both rainfall and snowfall, is a key component of the energy and water cycle 45 influencing the Earth’s climate system. Its measurement is not only fundamental in specifying the 46 current state of the distribution and intensity of precipitation that help define our climate, but also for 47 monitoring the changes in our climate. Precipitation is considered to be an essential global variable 48 (NASA 1988) and an Essential Climate Variable (GCOS 2010), and thus requires adequate 49 measurement. Fundamental to this must be high quality, long term observations at fine temporal and 50 spatial resolutions. Trenberth et al. (2003) emphasized the need to be able to assess and quantify the 51 changing character of precipitation through better documentation and processing of all aspects of 52 precipitation. In particular, Stephens et al. (2010) noted that precipitation is not well represented in 53 climate-scale models. Precipitation is also of great interest to a number of different scientific 54 disciplines beyond the atmospheric community, including the hydrological, oceanic, cryospheric, 55 environmental, ecological and biological communities. Not only is precipitation a critical component of 56 the Earth System, but also essential to life on Earth, impacting not only humanity, but also the natural 57 environment around us. Over land, precipitation is ultimately the source of all fresh water. The 58 monitoring and measurement of precipitation is of economic value for agriculture through agro- 59 businesses such as crop forecasting, water resource management, civil defense through mitigation of 60 droughts or floods, and through more benign economic returns through, for example, the removal of 61 particulate matter from the atmosphere (Thornes et al. 2010). 62 63 The measurement of precipitation (defined as deposition of water from the atmosphere in solid or 64 liquid form) might at first appear to be straightforward; however, precipitation is relatively rare, highly 65 variable, and consequently poorly monitored as an environmental parameter particularly on a global 66 basis. Instantaneously, precipitation occurs globally probably less than 1% of the time (Barrett and 67 Martin, 1981). When precipitation does occur, intensities may range from very light to very heavy; the 3 68 range of intensities for instantaneous precipitation is highly skewed towards lighter intensities. 69 Furthermore it has significant spatial and temporal variability, making it difficult to measure 70 satisfactorily; dense observational networks are necessary to adequately capture this variability, 71 particularly at fine temporal and spatial scales. Averaging over time and space generally results in 72 accumulated precipitation being more normally distributed and more representative (Bell et al. 1990); 73 climatological-scale accumulations require less dense networks, although these may not necessarily 74 faithfully capture small scale, extreme events or the variability over complex terrain. 75 76 Thus, the adequate measurement of precipitation is necessary at a number of scales and for a number of 77 users. For flash flood studies precipitation measurements are required at local, fine scales with rapid 78 access to the data (low latency) while for drought, longer term measurements will suffice, with less 79 stringent spatial, temporal and latency requirements. For climate studies the accuracy of the 80 measurements and the homogeneity of the data record are perhaps paramount over other criteria to 81 enable the assessment of the subtleties due to climate change. 82 83 Gauge numbers 84 The number of gauges cited in the literature varies somewhat. In their Catalogue of National Standard 85 Precipitation Gauges, Sevruk and Klemm (1989b) put the number of gauges worldwide at more than 86 150,000, while Groisman & Legates (1995) estimated the number of ‘different’ gauges to be as many 87 as 250,000. However, New et al. (2001) put the number closer to the figure of 150,000 stations of 88 Sevruk and Klemm. The figure was quantified by Strangeways (2003) who identified at least 123,014 89 monthly accumulation gauges (summarized in Table 1). These variations are largely dependent upon on 90 the criteria used to count the number of gauges; for example, some of these numbers will include all the 91 ‘stations’ that have existed and have provided some precipitation measurements at some time in their 4 92 observational record, while others will only report locations which currently return precipitation 93 measurements. Thus, while it is certain that many gauges exist, not all gauges have operated 94 continuously or simultaneously. 95 96 Not all gauge observations are available to the public or even to researchers. Those observations that 97 are available, are not necessarily available for all temporal samples (i.e. 3-hourly, daily, etc), or with 98 adequate data latency; flood monitoring and forecasting requires the timely delivery of data to be truly 99 useful, whereas climate application can accommodate longer data delivery times. The availability of 100 data from different countries/regions often depends upon the organization within the country, region or 101 locality. Often more than one agency within each country is tasked with the collection of rainfall data; 102 these agencies are not necessarily consistent from one country to the next. An additional and potentially 103 large number of gauge observations are available from commercial networks (e.g. water companies) 104 although such data may be deemed to be commercially sensitive and therefore access to such data is 105 often restricted. 106 107 Global meteorological data (including precipitation) is available through the World Meteorological 108 Organisation (WMO) Global Telecommunication System (GTS), collected from between 8,000 and 109 12,000 “first class” stations (WMO, 2011). The precipitation information contained with the SYNOP 110 report is collected for 3-hourly and daily periods at the fixed synoptic hours and distributed in near real 111 time, although the records for each station may not always be complete for an entire monthly record. 112 Figure 1 illustrates the coverage of these measurements by mapping the distance from each of the GTS 113 stations across the globe; it can be seen that the data coverage for near real-time data on a global scale 114 is relatively poor. While some regions such as Europe and eastern Asia (including Japan) have 115 reasonable coverage, elsewhere gauges are sparse. This means that applications such as flash flood 5 116 monitoring that require fine temporal and spatial resolutions generally rely upon gauge and radar 117 (where available) observations obtained from local or regional meteorological organizations, or 118 satellite-based infrared estimates (Arkin and Xie, 1989) 119 120 At the daily scale, the situation is somewhat better. A more comprehensive set of daily gauge data is 121 organized through the Global Precipitation Climatology Project (GPCP) at the Global Precipitation 122 Climatology Centre (GPCC; Becker et al. 2013) which provides perhaps the foremost repository of 123 global precipitation data derived from gauges. Access to existing data sets hitherto unavailable to the 124 GPCC has been improved through the WMO-implemented Global Terrestrial Network for Hydrology 125 (GTN-H) observing system since 2001. Although the data released by the GPCC is restricted to a 126 gridded product, it reveals the number of rain gauges operating across the globe that report information 127 on a regular and reliable basis. As of 2013 (2015) a total of 180 institutions contribute data to the 128 GPCC from about 85,000 (100,000) gauge locations that have provided observations at least once since 129 the start of the dataset in 1901. Initial daily and monthly products are available a few days after the end 130 of the integration period, with a more complete ‘monitoring’ product after about 8 weeks and full daily 131 and monthly products available after about 2 years. For this full, long-term or climatological analysis it 132 is critical to ensure continuous records of precipitation from any single station, consequently the GPCC 133 imposes a 10-year minimum constraint. This restricts the number of available stations as of 2013 134 (2015) to 67,298 (75,165) for the best month, or 67,149 (75,033) for the worst, or a total 65,335 135 (73,586) stations across all 12 months of the year (Becker et al. 2013; Schneider et al., 2015). Figure 2 136 shows the coverage of the GPCC gauge data. Most of Germany lies within 10 km of the nearest rain 137 gauge, while large areas of Europe, the US, eastern South America, India and the more populated 138 regions of Australia are less than 25 km from a gauge. Other regions with lesser, but still good 139 coverage include Turkey and Iran, parts of Africa (South Africa in particular) and the Andes in South 6 140 America. Some of the GTS stations ‘disappear’ in the GPCC dataset primarily due the fragmented 141 nature of their observational record. 142 143 A number of other key gauge data products exist that provide a greater range of precipitation products 144 at varying temporal and spatial resolutions. It should be noted that many of these data products utilize 145 the same gauge information as the GPCC product, rather than providing information from additional 146 gauges. Such global data sets include the CPC Gauge-Based Analysis of Global Daily Precipitation 147 (Xie et al. 2010) and the Global Historical Climatology Network (GHCN; Menne et al. 2012), both of 148 which provide daily gridded precipitation products derived from meteorological observations 149 worldwide. The number of available gauges varies considerably by year (and by region/year) with a 150 maximum (for precipitation observations) of just over 30,000 stations, about half of which are in the 151 US. The GHCN also collects information on snow depth from about 17,000 stations, again virtually all 152 in the US. The Climate Research Unit at the University of East Anglia gauge product (Mitchell and 153 Jones 2005) aims to provide a consistent precipitation data set exploiting historical precipitation 154 records. Regional data sets, such as the APHRODITE product (Yatagi et al. 2012) and the China 155 Gauge-based Daily Precipitation Analysis (CGDPA; Shen and Xiong 2016) are often able to obtain a 156 greater number of regional gauges through local sources. 157 158 It is therefore clear that the number of gauges used in creating precipitation products varies 159 considerably. The numbers of sub-daily rainfall gauge observations available in near real-time is small, 160 although more observations are available if the user is willing to wait longer for the data to become 161 available. Daily gauge accumulations, although hindered by non-uniform reporting times globally, 162 represent perhaps the greatest number of official data entries since this is in line with the WMO 163 recommendations and most easily implemented by the individual meteorological agencies. At longer 7 164 time scales the potential number of stations declines slowly, not least if a complete data record is 165 required since some stations might not report precipitation (including zero-rain) 100% of the time. 166 167 Gauge Representativeness 168 If the rain gauges alone are considered, the surface area of the orifices is surprisingly small. The most 169 common gauges, as noted in Table 1, provide a total surface area estimated to cover just 3,026 m2 from 170 123,014 gauges. Scaling the GTS and GPCC data sets using an average orifice size of 246 cm2 would 171 result in equivalent surface areas of about 295 m2 and 1,612 m2 respectively. For comparison, Table 2 172 provides the areas of pitches/courts/fields for common sporting activities; the comparisons between the 173 GTS and GPCC against the equivalent areas are illustrated in Figure 3. For the 3-hourly GTS data set, 174 assuming that the maximum number of gauges report data, an area just greater than that of the center 175 circle of a soccer pitch is actually measured; in reality less than half of the GTS stations regularly 176 report rainfall measurements. The GPCC gauges provide an area equivalent to about 4 basketball 177 courts. 178 179 However, fundamental to the measurement of precipitation using rain gauges is that they are accurate at 180 the location and are representative of their surrounding area.The ‘capture’ of precipitation, particularly 181 solid precipitation, by a rain gauge is largely affected by the wind-effect around the orifice, an effect 182 that is exacerbated with increased exposure (Duchon and Essenberg, 2001; Goodison et al, 1998), 183 together with losses or errors that may also arise from the mechanical construction of the gauge. 184 However, despite errors associated with rain gauges, they remain arguably the most accurate 185 instrument by which to measure rainfall. The measurement of snowfall is more difficult than the 186 measurement of rainfall due to nature of falling (and blowing) snow, the variety of snow gauges used 187 and the catchment (in)efficiencies of the gauges and is the focus of the WMO Solid Precipitation 8 188 Intercomparison Experiment (SPICE) project (Nitu and Wong, 2010b, ; Rasmussen et al, 2012). The 189 majority of these measurements are now made by automated systems (Nitu and Wong 2010a), 190 predominantly by weighing or tipping bucket gauges, the latter being poor at measuring snowfall 191 (Goodison et al. 1998). Despite the measurement accuracy for snowfall being strongly affected by the 192 wind due to the collector-snow particle flow dynamics, only about 28% of precipitation gauges are 193 equipped with shields to modify the air flow over the gauge, although most automated snow gauges are 194 heated in order to prevent snow accumulating on the rim or sides of the collector (Nitu and Wong, 195 2010a). While rainfall can be usually be measured to within 10-20% (Vuerich et al, 2009), wind-effects 196 may result in less than 25% of the snowfall being caught (Goodison et al. 1998). However, errors and 197 uncertainties associated with such precipitation measurements for manual gauges are reasonably well 198 understood and corrections (or quality control) can be applied. The SPICE project is currently 199 addressing corrections necessary for automatic gauges. 200 201 Spatially, at the very local scale, the gauge should at least represent the rainfall falling in its immediate 202 vicinity, over scales of a few metres and preferably a few kilometres. However, gauge measurements 203 have their limitations given the spatial and temporal variability of precipitation and the fact that gauges 204 are (small) point measurements. Standards set by the WMO (2008) are designed to ensure consistency 205 between gauge measurements to reduce some of the inherent errors, such as those caused by siting or 206 exposure. However, even under ideal situations the representativeness or auto-correlation length of 207 precipitation is surprisingly small; Habib et al. (2001) showed that for instantaneous precipitation over 208 the mid-western US the correlation coefficient between adjacent gauges fell to less than 0.5 just 4 km 209 away; similar results were found for frozen precipitation. Furthermore, this correlation length is 210 dependent upon the meteorology of the precipitation event and the local topography. Fortunately, 211 accumulating precipitation over time, increases the correlation length (Bell et al. 1990); over longer 9 212 periods, the gauges become more representative of the regional precipitation regime. Although many 213 schemes exist for the interpolation of precipitation, care is needed since the same interpolation scheme 214 applied to instantaneous or monthly precipitation data could produce undesired results: Indeed, the 215 interpolation of instantaneous gauge data should be avoided where possible due to the inherent 216 heterogeneity of precipitation at fine temporal and spatial scales. 217 218 Considering the representativeness of gauges on a global scale, Figure 4 illustrates the area of the Earth 219 within the defined distances from the GTS and GPCC gauge locations, divided into four regions, ocean 220 or land and 60°-polewards or 60°S-60°N. It is clear that the vast majority of the Earth’s surface closest 221 to gauges are (not surprisingly) concentrated over the land areas between 60°S-60°N, with relatively 222 few gauges over land polewards of 60°. Over the oceans only a very small area is within 100 km of a 223 gauge, and most of this area would be deemed ‘coastal waters’. Considering the GPCC data globally, 224 only 1.6% of the Earth’s surface lies within 10 km of a rain gauge, although 5.9% lies within 25 km; 225 over 60°S-60°N land areas this improves to 6.5% and 23.0% respectively. This contrasts with less than 226 4% of the Earth’s oceans lying within 100 km of a gauge. 227 228 Filling the gaps 229 It is clear that gaps exist within the currently available gauge networks over the various temporal scales 230 which require additional information if the representativeness of the precipitation measurements are 231 sufficiently adequate to meet user requirements. Despite significant progress having been made in 232 addressing some of the larger data gaps resulting from non-availability of regional gauge data sets, it is 233 also clear that not all existing rain gauges that could be used are currently exploited. The gauges 234 incorporated into the GPCC database derive from meteorological agencies which adhere to the 235 requirements laid down by the WMO to ensure consistent measurements between different sites and 10